Solving Inverse Problems With Deep Neural Networks -- Robustness Included?